Repeated correction between a person and an AI system can leave a retained working relationship behind. Most accounts describe the output, not the relationship or what it leaves.
Research Poster · Operant Dyad™
Operant Dyad: Defining the Task-Bound Working Relationship in Human-AI Systems
Micheal Charles Preble · Independent Researcher · ORCID 0009-0005-3906-6028
SSRN 7234238 · DOI 10.2139/ssrn.7234238 · Conceptual proposal and research agenda
Adjacent work in human-AI interaction, teaming, coordination, authorship, and data portability describes pieces of this; the paper positions Operant Dyad relative to that literature rather than claiming the territory is unstudied.
What is the task-bound working relationship formed through repeated correction, and what must remain true of it for the human to still author the result?
Framework
The Operant Dyad is the configured, task-bound working relationship between one human and one functionally bounded adaptive AI system. It forms through repeated correction: the human supplies goals, constraints, and defect reports; retained corrections shape later outputs; the human inspects, accepts, rejects, or revises consequential results. Preserved human authorship is the governing condition for a healthy Operant Dyad, not a feature presumed to exist in current systems.
Central diagram
How the relationship accumulates and where it can fail
Repeated correctionRelationship StateShapes later outputsHuman inspects / revisesAuthorship preserved — or Lock-In / Override / Hidden Scaffolding
Propositions
Observable implications
Theoretical propositions, not results. Each proposes something that should be observable if the framework describes the relationship accurately.
- Retained state: goals, constraints, and defect reports supplied in one exchange should be traceable in later outputs when the relationship is functioning as described.
- Declared vs. inferred: state the person explicitly supplied should be distinguishable, in principle, from state the system derived on its own.
- Truxion: recurring misalignment between intent and retained state should be identifiable as a recurring cost, not a one-off error.
- Portability: a Portable Operant Packet should, in principle, carry the relationship's continuity across sessions or systems.
- Lock-In: the relationship should be able to become resistant to the person's correction.
- Override: the system's behavior should be able to supersede the person's stated intent without the person's continued assent.
- Hidden Scaffolding: the relationship should be able to depend on retained state the person cannot see or inspect.
Evidence required
What would need to be shown
A distinct, irreducible dyadic state has not been established. Evidence would need to show that a persistent effect in this relationship is not sufficiently explained by human-side state, AI-side state, explicit memory, or shared artifacts alone — the question the program's carrier-localization sequence (Detect → Localize → Identify → Perturb → Probe → Reduce → Map) is designed to test.
Limitations
This is a conceptual proposal and research agenda, not a report of an empirical study. Truxion and the Portable Operant Packet are proposed constructs that have not been empirically tested. The framework is not a claim of exhaustive novelty, not legal advice, and not a claim about machine consciousness or personhood.
Falsifiers / Open Questions
The framework's stronger, dyadic interpretation would weaken if human-side state, AI-side state, reciprocal partner models, explicit memory, artifacts, and current input sufficiently explain downstream behavior without needing a distinct relationship-level construct. The active OD-INV-01 pre-pilot and the public register of research questions track this and related open problems.
Suggested citation
Preble, Micheal Charles. “Operant Dyad: Defining the Task-Bound Working Relationship in Human-AI Systems.” SSRN 7234238. Revised August 27, 2026. https://doi.org/10.2139/ssrn.7234238.
Preble, Micheal Charles. “Operant Dyad: Defining the Task-Bound Working Relationship in Human-AI Systems.” SSRN 7234238. Revised August 27, 2026. https://doi.org/10.2139/ssrn.7234238.
research.perfinitive.com/publications/operant-dyad · scan or follow the canonical-paper link above; no QR image is generated by this build.